The Pixel Anomaly Detection Tool: a user-friendly GUI for classifying detector frames using machine-learning...

Area of Science:

• Crystallography
• Imaging Science
• Data Science

Background:

• X-ray free electron laser (XFEL) data collection faces challenges like continuous sample delivery and novel detector technologies.
• Data artefacts from XFEL experiments can hinder accurate structure-factor amplitude determination for serial crystallography and single-particle imaging.

Purpose of the Study:

• To develop and present a novel data-classification tool for XFEL experimental data.
• To enable accurate structure-factor amplitude determination by effectively sorting and cleaning experimental data.

Main Methods:

• Implementation of a machine-learning (ML) based data-classification tool with various algorithms.
• Training the ML model using manual user sorting or intensity distribution profile fitting.
• Integration into a user-friendly graphical user interface (GUI) supporting common XFEL detectors, file formats, and software.

Main Results:

• The tool successfully sorts data, removing unwanted artefacts detrimental to structural analysis.
• Supervised learning approach allows novice users to perform data sorting and hit finding without coding.
• The modular design ensures expandability to other X-ray sources and detectors.

Conclusions:

• The developed ML tool effectively addresses data artefact challenges in XFEL experiments.
• It enhances the accuracy of structure-factor amplitude determination in serial crystallography and single-particle imaging.
• The tool democratizes data analysis for XFEL users, simplifying routine tasks and improving experimental outcomes.